Ready for review: new std.uni

[Dmitry Olshansky]

12-Jan-2013 00:38, David Nadlinger пишет:
> On Friday, 11 January 2013 at 19:31:13 UTC, Dmitry Olshansky wrote:
>> The code, including extra tests and a benchmark is here:
>> https://github.com/blackwhale/gsoc-bench-2012
>
> Phew, that repository is quite a hefty checkout with all the example data.
>
> Just for the fun of it, a completely irreproducible, unscientific
> benchmark (sorry, Andrei!) run on a Core i7 M 620 under Linux x86_64:
>
> DMD 2.061
> ---
> $ rdmd -O -release -noboundscheck -inline char_class.d
[snip]

I'd rather use dewiki or arwiki these give the more coverage otherwise 
you just check the presence of ASCII special casing (only isAlpha and 
isWhite have it).

> Time to build dchar array: 2702 ms
>
> Baselines
>              noop [enwiki],   217.335, 888.99 M/s
>     new-std-alpha [enwiki],   1197.49, 161.34 M/s
>      new-std-mark [enwiki],   6374.73, 30.31 M/s
>       new-std-num [enwiki],   6290.89, 30.71 M/s
>       new-std-sym [enwiki],   6500.93, 29.72 M/s
>     new-std-white [enwiki],   822.245, 234.98 M/s
>   combining class [enwiki],   5948.63, 32.48 M/s
>         inv-alpha [enwiki],     10290, 18.78 M/s
>          inv-mark [enwiki],    9229.2, 20.93 M/s
>           inv-num [enwiki],   9046.48, 21.36 M/s
>           inv-sym [enwiki],   9843.59, 19.63 M/s
>
> Tries of level 1
>        trie-alpha [enwiki],   2179.17, 88.66 M/s
>         trie-mark [enwiki],   1912.54, 101.02 M/s
>          trie-num [enwiki],   1924.92, 100.37 M/s
>          trie-sym [enwiki],   1911.22, 101.09 M/s
>
> Tries of level 2
>        trie-alpha [enwiki],   4524.33, 42.70 M/s
>         trie-mark [enwiki],   4276.92, 45.17 M/s
>          trie-num [enwiki],   4308.42, 44.84 M/s
>          trie-sym [enwiki],   4298.68, 44.95 M/s
>
> Tries of level 3
>        trie-alpha [enwiki],   6407.64, 30.15 M/s
>         trie-mark [enwiki],    6150.8, 31.41 M/s
>          trie-num [enwiki],   6095.96, 31.69 M/s
>          trie-sym [enwiki],    6253.7, 30.89 M/s
>
> Tries of level 4
>        trie-alpha [enwiki],   8329.66, 23.20 M/s
>         trie-mark [enwiki],   7993.11, 24.17 M/s
>          trie-num [enwiki],   8040.94, 24.03 M/s
>          trie-sym [enwiki],   8033.07, 24.05 M/s
> ---



> LDC (2.061 merge branch):
> ---
> $ rdmd --compiler=ldmd2 -O3 -release -noboundscheck -inline char_class.d
> enwiki-latest-all-titles-in-ns0

[snip]

> Baselines
>              noop [enwiki],   211.516, 913.44 M/s
>     new-std-alpha [enwiki],   612.324, 315.53 M/s
>      new-std-mark [enwiki],   577.134, 334.77 M/s
>       new-std-num [enwiki],   600.996, 321.48 M/s
>       new-std-sym [enwiki],   577.433, 334.60 M/s
>     new-std-white [enwiki],   381.535, 506.40 M/s
>   combining class [enwiki],   517.231, 373.54 M/s
>         inv-alpha [enwiki],   8181.25, 23.62 M/s
>          inv-mark [enwiki],   6938.09, 27.85 M/s
>           inv-num [enwiki],   6347.71, 30.44 M/s
>           inv-sym [enwiki],    6930.7, 27.88 M/s
>
> Tries of level 1
>        trie-alpha [enwiki],   275.741, 700.69 M/s
>         trie-mark [enwiki],   274.972, 702.65 M/s
>          trie-num [enwiki],   274.558, 703.70 M/s
>          trie-sym [enwiki],   277.255, 696.86 M/s
>
> Tries of level 2
>        trie-alpha [enwiki],   453.453, 426.08 M/s
>         trie-mark [enwiki],   452.912, 426.59 M/s
>          trie-num [enwiki],   455.004, 424.63 M/s
>          trie-sym [enwiki],   453.492, 426.04 M/s
>
> Tries of level 3
>        trie-alpha [enwiki],   577.442, 334.59 M/s
>         trie-mark [enwiki],   579.054, 333.66 M/s
>          trie-num [enwiki],    577.95, 334.30 M/s
>          trie-sym [enwiki],   577.566, 334.52 M/s
>
> Tries of level 4
>        trie-alpha [enwiki],   778.261, 248.26 M/s
>         trie-mark [enwiki],   778.943, 248.04 M/s
>          trie-num [enwiki],   778.352, 248.23 M/s
>          trie-sym [enwiki],   779.442, 247.88 M/s
> ---
>
> Haven't had a look at what is going on there yet – it could well be that
> some measurement loop is optimized out altogether.
>

You can print total counts after each bench, there is a TLS varaible 
written at the end of it. But anyway I like your numbers! :)

> David


-- 
Dmitry Olshansky